grilllisted
Install: claude install-skill orobsonn/claude-harness
# Grill — interrogate an idea until it becomes a PRD
The operator shows up with an idea that only exists in his head. This skill turns it into a
**written PRD** by asking questions until nothing that could change a requirement is still open.
Announce at start (pt-br): "Vou te entrevistar até essa ideia virar um PRD."
All reasoning and identifiers stay in English. Every message to the operator is **pt-br,
product-language**. The PRD section headings are pt-br on purpose — the PRD is an operator-facing
artifact.
---
## Where this sits — and what it is not
- **Before the pipeline.** It runs *before* triage, before any spec, before any plan. Its output is
the input the pipeline later consumes.
- **NOT the entry gate.** `triaging-requests` owns the first action of every session. This skill
never claims that slot, never says "run me first", never stamps a marker, never calls
`classify.mjs` or `mark.mjs`, and gates nothing. It is entered only when the operator asks for it
or accepts a suggestion to use it.
- **NOT a source of truth.** *(the single most important line in this file)* Everything the grill
produces enters the pipeline later as **ordinary input** — issue text. It is fully attackable and
re-questionable by the planner, `plan-reviewer`, `adversary`, and `compliance` downstream. **Never
record grill output as a "locked decision" that downstream roles are obliged to defend.** That
status belongs to `brainstorming`'s in-pipeline elicitation, not to this